Presentation is loading. Please wait.

Presentation is loading. Please wait.

Structured Forests for Fast Edge Detection

Similar presentations


Presentation on theme: "Structured Forests for Fast Edge Detection"— Presentation transcript:

1 Structured Forests for Fast Edge Detection
Piotr Dollár and Larry Zitnick

2 Let the data speak. what defines an edge?
Brightness Color Texture Parallelism Continuity Symmetry … Let the data speak.

3 1. Accuracy 2. Speed

4 I. data driven edge detection

5 edge detection as classification
{ 0, 1 } Hard! positives Supervised Learning of Edges and Object Boundaries CVPR 2006, Piotr Dollár, Zhuowen Tu, Serge Belongie

6 edge have structure

7 sketch tokens Sketch Tokens, CVPR Joseph Lim, C. Zitnick, and P. Dollár

8 random forests

9 upgrading the output space
dimensionality 2 { 0, 1 } { … } 151 2256

10 II. structured edge learning

11 structured forests Structured Class-Labels in Random Forests for Semantic Image Labelling, ICCV 2011, P. Kontschieder, S. Rota Bulò, H. Bischof, M. Pelillo

12 tree training

13 node training low entropy split  high entropy split 

14 how to train? ? ? good split  bad split 

15 minimize entropy cluster ? ?

16 III. structured edge detection

17 structured forests

18 sliding window detector

19 sliding window detector
Comparison to storing only single pixel prediction Noisy, no coherent structure Would need far more trees! This is the secret to our speed / accuracy. pixel output  structured output 

20 multiscale detection

21 multiscale detection + + ½ x 1 x 2 x

22 IV. results

23 SS=single-scale MS=multi-scale T=# trees
SS T=1 ODS = 0.72 60Hz SS T=4 ODS = 0.73 30Hz MS T=4 ODS = 0.74 6Hz gPb ODS=.73 FPS ≈ 1/240 Hz SS=single-scale MS=multi-scale T=# trees gPb ODS=.73

24

25

26

27

28 thanks! source code available online


Download ppt "Structured Forests for Fast Edge Detection"

Similar presentations


Ads by Google